Distributed Training of Large Graph Neural Networks with Variable Communication Rates

We train GNNs on large scale graphs in a distributed way using variable compression for the communications between machines.